Report on Correcting Tropical Biases Meeting

Prepared by Edwin K. Schneider

 

Summary

 

The Correcting Tropical Biases Meeting was held at the College Park Sheraton (adjacent to COLA) in Calverton, MD on September 8-9, 2005. The objective of the meeting was to initiate a program of experimentation to understand the causes of and to alleviate tropical biases in coupled general circulation models (CGCM). The meeting focused on biases in the tropical Pacific upper ocean temperature, precipitation, and surface wind stress in the annual mean, annual cycle, and interannual variability. Representatives of modeling and model diagnostic centers attended, as well as university based theoreticians and modelers. Seven experiments were decided on for proof-of-concept studies to be carried out by lead investigators, and groups interested in pursuing these experiments with their models were organized. The preliminary experiments will be carried out and written reports will be disseminated by February 2006. Those experiments that appear promising will be refined and carried out by the members of the respective groups. A workshop to discuss results and implications is anticipated to be associated with the 2006 CCSM Workshop.

 

Logistics

 

Venue: College Park Sheraton, Calverton, MD

Web page: www.iges.org/ctbm05

Organizing committee: Ed Schneider (George Mason University/COLA) chair; Ed Sarachik (University of Washington); Bill Large (NCAR); Jim Kinter (COLA)

Funding: NSF, NOAA, NASA, DOE

Attendees: see Appendix 1

 

Exchange of Information

 

The discussion and proposed experiments focused on the simulation biases in the tropical Pacific Ocean, although there is of course great interest in the Indian and Atlantic Oceans. Problem areas in the Pacific, which were described in detail at the First Tropical Biases Workshop, are:

1. Annual mean SST – cold tongue extends too far west, warm pool too cold, warm bias near S. American coast, double ITCZ, heat sources too far east.

2. Annual cycle of SST – warmest too late in the year, coldest too early, near the S. American coast where the amplitude is the largest. Indicative of too strong a semi-annual cycle.

3. ENSO – generally not well done. Amplitude biased towards western Pacific, wind stress anomalies too narrow and too far west. ENSO period short.

 

The experiences of several modeling groups in attempting to correct the biases were described and discussed. These are categorized and summarized below.

  1. Resolution/numerics
    1. Increased horizontal resolution – CCSM and GFDL see some improvement in E. Pac. SST annual mean bias with increased AGCM horizontal resolution.
    2. Increased vertical resolution reduces biases – cold bias, weak ENSO and MJO (NCEP); double ITCZ (GFDL, resolution increased near lower boundary).
    3. Numerical representation of dynamical core grid (finite volume) vs. spectral

1.     GFDL - FV gives doubled ENSO amplitude (too large), eliminates equatorward shift in subtropical jets, worse tropical simulation.

2.     CCSM – FV reduces trades, heat flux.

  1. Parameterizations
    1. Convective momentum transport (CMT)

1.     GFDL – lengthens ENSO period, improves (broadens) structure, no impact on double ITCZ, reduces MJO

2.     CCSM – uncoupled simulations with CAM showed no effect with GFDL parameters. Must increase damping by factor of 5 to see an effect, which is not beneficial in that high frequencies were strongly damped without reducing biases.

    1. Convective heating

1.     COLA – lowering/raising cloud base reduces/increases equatorial easterly wind stress.

2.     Modifying vertical profile of convective heating impacts surface winds.

3.     Eliminating convective heating parameterization (deep and shallow) in three T31L18 AGCMs (COLA, CCM3, ECHAM4), with all condensation processes handled by large scale parameterization reduces double ITCZ bias, model differences, improves MJO at the expense of other problems (7°C colder tropical troposphere, unrealistic diurnal variability).

4.     GFDL - moist convective adjustment in place of RAS. Reduces double ITCZ, increases ENSO damping by heat flux, better MJO.

5.     Rainfall reevaporation

1.     Goddard - reevaporation of precipitation in the convective parameterization provides a convenient and useful tuning parameter for AGCM. Increasing reevaporation reduces double ITCZ bias. Strong reevaporation leads to a mid-tropospheric wet bias.

2.     GFDL – no sensitivity.

    1. Ocean mixing

1.     Viscosity - reduced horizontal viscosity in ocean gives stronger tropical instability wave E. Pac. reduces cold bias in the east without degrading the west (CCSM).

2.     Background diffusivity - too large gives warm S. American coast, cool equator by affecting the thermocline structure, temperature of the upwelled water, position of upwelling. As diffusivity is made larger, more of the upwelling is along equator and less along the coast. Time scale of 10s of years (IPRC).

    1. Ocean color – included by CCSM, GFDL.
    2. Cloud - reduce too strong cloud/radiative feedbacks by shortening the cloud erosion time scale (GFDL).
  1. Coupling
    1. Diurnal cycle – including diurnal cycle of solar radiation in coupling to ocean leads to warmer tropics by 1°C (CCSM). Included by GFDL by coupling every 2 hours.
    2. Ocean surface currents – including in coupling helps (done already in several models including CCSM, GFDL).
  2. Flux override experiments
    1. Wind stress E. Pacific - replacing coupled model winds by AMIP or measured winds corrects annual cycle biases, but only if both the u and v components are replaced, not when only one component is replaced (CCSM).
    2. Heat flux E. Pacific

1.     Shortwave accounts for almost half the bias off Peru, and winds much of the rest in CCSM2, but this has not yet been tested in CCSM3.

2.     Cloud substitution gives correct annual cycle (GFDL).

    1. Ocean

1.     Restoring ocean along S. American coast produces beneficial effects that spread to the central Pacific. Does not affect W. Pacific double ITCZ. Coastal biases and western extension of the cold tongue biases may be distinct problems that are due to different mechanisms (CCSM).

2.     Restoring ocean structure to observed everywhere – little reduction in ENSO biases found (COLA), but gives some improvement in SI prediction (Chang).

    1. Anomaly coupling – COLA/MOM3 large impact on ENSO, CAM3/MOM3 no impact on ENSO (COLA).
    2. Multi-model interactive coupling – COLA AGCM and CAM coupled simultaneously to the same OGCM. One supplies heat flux and the other momentum flux. Best hybrid simulation using COLA wind stress, CAM heat flux. Also better than simulation of either model separately.

 

Mechanisms

  1. Role of the meridional structure of the zonal wind stress anomalies (from intermediate coupled models).
    1. A wider zonal wind stress anomaly should lead to a longer ENSO period by exciting slower off-equatorial Rossby waves (Kirtman) al la the Delayed Oscillator mechanism. An analogous argument applies to the Recharge Oscillator.
    2. Different wind stress anomaly meridional widths excite ENSO modes with different periods (Jin).
    3. CMT may lengthen ENSO period by broadening the meridional scale of the surface wind anomalies.
  2. Poor simulation of stratus and poorly resolved Andes may be responsible for annual cycle bias near S. America. Better vertical resolution (for stratus) and horizontal resolution (for Andes) may then improve this bias.
  3. Annual cycle of a slab mixed layer ocean “without” atmospheric or oceanic dynamics would have biases very similar to those found in CGCMs (Schneider). The most important wind effect w.r.t. the annual cycle is on the evaporation (Neelin).
  4. Relationship of mean state and annual cycle to ENSO.
    1. Phase locking of ENSO to annual cycle may be poorly simulated if annual cycle is poor, say if event growth is affected by instability of the annually varying basic state. GFDL has phase locking to a semiannual period, contrary to observations.
    2. Cold tongue bias affects ENSO structure. Precipitation anomalies occur with double ITCZ structure.
    3. Simulated ENSO variability ranges from “weak” to “realistic” to “strong,” but there is little understanding of why.
  5. Westerly wind bursts and ENSO (Tziperman). Westerly wind bursts in the equatorial western Pacific are associated with phenomena like paired cyclones, MJO. They are not purely stochastic, but are coupled to the annual cycle, and their occurrence depends on SST. There are 2-8 per year.
    1. Compared to stochastic forcing, WWB-like forcing coupled to SST produces twice ENSO amplitude in CZ model (stable regime).
    2. Low frequency component is important.
    3. WWB contribute preferentially to westerly wind stress. If they are not simulated well (due to poor spatial resolution for example), there might then be too strong easterlies.
    4. ECMWF finds improvement in predictions when parameterized WWB are included.
  6. SH moisture convergence in the incorrect part of the double ITCZ is mostly due to transients, NH mostly from time mean in GMAO model. This suggests that the double ITCZ may be due to too strong a feedback between convection and boundary layer.

 

Guiding principles

  1. Eastern and western tropical Pacific biases may involve different mechanisms and require different cures.
  2.  Development and experimentation should be done only in the coupled atmosphere-ocean model framework, as opposed to the more common strategy of coupling the “best” atmosphere and ocean components, where these are developed without reference to the coupled biases.
  3. Biases in the mean and annual cycle develop rapidly (within seasons to years), so only relatively short coupled integrations are required to examine the sensitivities of these biases.

 

Experiments

Some of the suggested experiments generated interest in several of the modeling groups. The experimental designs were left at an early stage, with details unresolved. It was decided to designate a lead for each experiment, who would attempt to perform it in a preliminary manner, and who would report within four months to the other groups on details of experimental design, feasibility, technical issues, and hopefully results. The seven ideas for experiments are described below, along with some motivation. The leads and those expressing interest in pursuing the experiments are listed in Appendix II. 

 

1.  Add estimated fluxes due to westerly wind bursts to fluxes provided to the ocean in the Western Pacific in CGCMs. The westerly wind burst parameterization should be a function of the SST, and hence the annual cycle of SST. Both mean effects (due to “westerly” wind bursts) and annual cycle of occurrence may be important. The mean westerlies could alleviate the easterly bias in the western Pacific zonal wind stress and the SST cold bias in that region. The dependence on SST could lead to changes in the ENSO properties, e.g. amplitude.

 

2. Explore sensitivity of AGCM and coupled simulations of the ITCZ/SPCZ to rainfall re-evaporation. Since the double ITCZ is sensitive to rainfall reevaporation (GMAO), it may be possible to reduce the double ITCZ bias in other AGCMs, and possibly to reduce the equatorial cold bias as well.

 

3. “Correct” the temperature of upwelling waters in the ocean in coupled simulations.

3a) Equatorial Pacific (away from coasts)

3b) Tropical Pacific near S. American coast

If the warm bias in the E Pacific and equatorial cold bias in the E/W Pacific are caused by biases in the structure of the thermocline (e.g. too diffuse) and/or problems in the diffusive heating parameterization (e.g. too strong), then the outcome of correcting these problems in the context of the CGCM can be examined by correcting the temperature of the upwelling waters. One way to do this is to modify the vertical advection operator, so that, for example, the temperatures used in this operator are displaced vertically from the vertical velocity. That is, a parcel could advect temperatures displaced, for example, 10m above the velocity level.  A more consistent alternative approach is to reduce the background diffusivity in the OGCM, which can also be expected to modify the current distribution.

 

4.  Suppress deep convection in regions of incorrect double ITCZ (SE Pacific). A number of methods were discussed to artificially suppress moist convection. Even in the most extreme case, with convection not parameterized, the AGCM produces stable results that are physically consistent in terms of the large scale budgets. In this experiment, deep moist convection will be suppressed in the E. Pacific south of the equator in the CGCM, eliminating the double ITCZ. The hypothesis is that this may reduce the equatorial easterly bias, thereby reducing the equatorial cold bias. In addition, biases in the dynamical structures associated with ENSO, such as the structures of precipitation and wind stress anomalies, could be reduced, leading to improved ENSO simulations.

 

5. Increase low level (below 500m) vertical resolution in AGCM. Increased vertical resolution in the boundary layer has been found to improve the double ITCZ bias.

 

6. Examine AGCM+mixed layer ocean response to warming of the tropical troposphere (without corresponding surface warming). The idea is to stabilize the atmospheric vertical temperature profile of the atmosphere as seen by the deep convection parameterizations, with values similar to those that occur in simulations with increasing CO2. However the warming of surface temperature will not be included. This will stabilize the atmosphere to moist convection and allow comparison of moist convective feedbacks on the moisture budget and atmospheric dynamics.

 

7. Relate specific initial errors in AGCM/CGCM simulations to biases. There is significant information on tropical biases in the initial tendencies of models. In the atmosphere, this information may be apparent at the first time step of a simulation, while in the CGCM the first month may be the appropriate initial period. The plan is to collect and analyze in detail the initial tendencies, especially from the heat budget, in control runs of CGCMs. The diagnostic intervals of every half hour for the 1st day and every day for the first 100 were suggested. If these can be related simply to the eventual biases, then the process of testing bias corrections can be made more efficient. Additionally, the structure of the initial tendencies in the budget equations may give clues as to how to correct the biases.

 

Diagnostics

  1. Examine output of CGCM control runs to see if WWB statistics are simulated reasonably.
  2. A diagnosis of the upper ocean heat budget in the cold tongue based on a simplified model of heat flux balancing upwelling was proposed (Jin). He was requested to make a detailed specification of the quantities needed from the models.
  3. A list of special diagnostics that have been found or were thought could be useful to examine w.r.t. tropical biases was developed. The diagnostics were thought to be most useful if  data were made available as netcdf datasets in IPCC format for examination by other groups and investigators. The list follows.
    1. Atmospheric heating profiles, net and components
    2. Vertical motion profiles in the atmosphere and upper ocean
    3. Atmospheric vertical velocity variance and 850 mb
    4. Ocean surface heat flux as a function of SST
    5. High frequency wind variability in the western Pacific (WWB)
    6. Correlation of SST with surface radiation (esp. cloud related), heat budget of the cold tongue
    7. Convective vs. large scale precipitation (already standard)
    8. ENSO metrics (GFDL Ferret package might be used/adapted to analyze these).

                                                    i.     Nino3, Nino3.4, Nino4, lat/lon SST variance

                                                  ii.     Seasonal cycle of SST variance

                                                iii.     Western extent and horseshoe pattern sign reversal

                                                iv.     Extratropical response

                                                  v.     SSTA (Nino3.4) vs. zonal wind stress

                                                vi.     SSTA vs. precip.

                                              vii.     SSTA vs. SLP

                                            viii.     SSTA vs. heat flux components (SW, LW, LHF)

                                                 ix.     Measures of duration of events, such as global maps of one month autocorrelations of SST

                                                   x.     Depth of 20°C isotherm in ocean

           

 

 

Actions

 

The lead investigators are taking responsibility for carrying out initial versions of the experiments in a timely manner, and for refining the experimental design. Schneider will continue to coordinate the activities initiated at the meeting. He will solicit, collect, and disseminate reports on progress with these experiments in February 2006. A web page will be maintained at COLA to provide a facility for accessing and communicating results. A follow-up meeting is anticipated in conjunction with the 2006 CCSM Workshop for discussion of the results and for planning further activities.

 

 

Appendix 1:

 

An alphabetical list of attendees and affiliation follows

 

Krishna Achutarao                  PCMDI            achutarao1 AT llnl . gov        

Julio Bacmeister                     GSFC              Julio . Bacmeister . 1 AT gsfc . nasa . gov     

Magdelena Balmaseda            ECMWF          Magdalena . Balmaseda AT ecmwf . int        

Anjuli Bamzai                        DOE                Anjuli . Bamzai AT science . doe . gov

Marcelo Barreiro                    Princeton         barreiro AT princeton . edu

Michela Biasutti                      LDEO              biasutti AT ldeo . columbia . edu       

Chris Bretherton                      UWash             breth AT atmos . washington . edu

Jim Carton                               UMd                carton AT atmos . umd . edu   

Ping Chang                              Texas A&M    ping AT tamu . edu

Gokhan Danabasoglu               NCAR             gokhan AT ucar . edu

Jay Fein                                   NSF                 jfein AT nsf . gov

Ryo Furue                                IPRC               furue AT hawaii . edu

Peter Gent                               NCAR             gent AT cgd . ucar . edu

Wayne Higgins                        NCEP              wayne . higgins AT noaa . gov

Bohua Huang                           GMU/COLA    huangb AT cola . iges . org

Ming Ji                                    NOAA             ming . ji AT noaa . gov

Fei-Fei Jin                               FSU                 jff AT met . fsu . edu

Jim Kinter                               COLA              kinter AT cola . iges . org

Ben Kirtman                            GMU/COLA    kirtman AT cola . iges . org

Barry Klinger                          GMU               klinger AT cola . iges . org

Bill Large                                NCAR             wily AT ucar . edu

David Legler                           CLIVAR          legler AT usclivar . org

Vasu Misra                              COLA              misra AT cola . iges . org

Raghu Murtugudde                  UMd                ragu AT essic . umd . edu

David Neelin                           UCLA              neelin AT atmos . ucla . edu

Sumant Nigam                         UMd                nigam AT atmos . umd . edu

Hua-Lu Pan                             NCEP              Hualu . Pan AT noaa . gov

Jerry Potter                              PCMDI            potter2 AT llnl . gov  

Phil Rasch                               NCAR             pjr AT ucar . edu

Kelvin Richards                      IPRC               rkelvin AT hawaii . edu

Tony Rosati                             GFDL              Tony . Rosati AT noaa . gov

Ed Sarachik                             U. of Wash.     sarachik AT atmos . washington . edu

Paul Schopf                             GMU               schopf AT cola . iges . org

Ed Schneider                           GMU/COLA    schneide AT cola . iges . org

J. Shukla                                  GMU               shukla AT cola . iges . org

Max Suarez                             GSFC              Max . J . Suarez AT gsfc . nasa . gov

Shan Sun                                  GISS                sun AT venus2 . giss . nasa . gov

Eli Tziperman                         Harvard           eli AT eps . harvard . edu

Roxana Wajsowicz                 U. of Md          roxana AT atmos . umd . edu  

Wanqiu Wang                          NCEP              Wanqiu . Wang AT noaa . gov

Yuqing Wang                           IPRC               yuqing AT hawaii . edu

Andrew Wittenberg                 GFDL              Andrew . Wittenberg AT noaa . gov  

Zhaohua Wu                            COLA              zhwu AT cola . iges . org

Pingping Xie                           NCEP              Pingping . Xie AT noaa . gov

Yan Xue                                  NCEP              Yan . Xue AT noaa . gov

Jin-Yi Yu                                UC Irvine        jyyu AT uci . edu        


Appendix II: Planned Experiments and Participants

           

1.  Add estimated fluxes due to westerly wind bursts to fluxes provided to the ocean in the Western Pacific.

            Lead: Tziperman

            Participants: CCSM (Large), GFDL (Rosati), Murtugudde, ECMWF(Balmaseda)

 

2. Explore sensitivity of AGCM and coupled simulations of the ITCZ/SPCZ to rainfall re-evaporation.

            Lead: GMAO (Bacmeister)

            Participants: COLA (Kirtman), CCSM (Rasch), GFDL (Rosati), PCMDI (Potter)

 

3. “Correct” the temperature of upwelling waters in the ocean in coupled simulations.

            3a) Equatorial Pacific (away from coasts)

            Lead: Schopf/Klinger

            Participants: CCSM (Danabasoglu),  IPRC (Richards), GFDL (Wittenberg)

            3b) Equatorial Pacific near S. American coast

            Lead: Schopf/Klinger

            Participants: CCSM (Danabasoglu), GISS (Sun), IPRC (Richards)

 

4.  Suppress deep convection in regions of incorrect double ITCZ (SE Pacific).

            Lead: CCSM (Bretherton)

            Participants: IPRC (Wang), COLA (Schneider), GMAO (Bacmeister)

 

5. Increase low level (below 500m) vertical resolution in AGCM.

            Lead: GFDL/CCSM (Rosati/Rasch)

            Participants: NCEP (Pan), ECMWF (Balmaseda), GMAO (Bacmeister), COLA (Misra)

 

6. Examine AGCM response to warming of the tropical troposphere (without corresponding surface warming).

            Lead: Neelin

            Participants: GMAO (Bacmeister, Suarez), CCSM (Rasch), IPRC (Richards), GFDL (?)

 

7. Relate specific initial errors in AGCM/CGCM simulations to biases.

            Lead: GMAO (Suarez)

            Participants: NCAR (?), IPRC (Potter), GFDL (Rosati), NCEP (Pan)